Research On Software Cost Estimation Model

With the development of computer and software industry, the scale of thesoftware is constantly expanding, and its complexity is constantly increasing, softwaregradually became the most expensive part of the entire computer system. However,since the sixties of last century, the typical characteristics of the software crisis tobudget overruns, quality defects and project delays continue to emerge.Therefore, toestimate the cost of the software project becomes particularly important.At present, various estimation methods have also born one after another, played acrucial role in their respective field of application, including analogy, expert andCOCOMO model. In this paper, we uses the statistical regression analysis, it hasstrong objectivity and a more mature theoretical system of support, could set toextract the important information during the complex historical data, make full use ofthese data to complete the software project cost estimation and forecasts.By the research of the theoretical basis of and algorithm theory of the regressionanalysis method, separate the core idea of monadic regression from the multipleregression、stepwise regression、ridge regression and weighted regression, and usethem to their applicable estimation stage during the process of the Software CostEstimation. Before the establishment of software cost estimation model, first to studythe influencing factors of software cost estimation, and eventually come to12impactfactors of the regression model should be introduced by the Analytic HierarchyProcess. On the basis of the above work, based on the method of regression analysisto construct the general process of software cost estimation, and the correspondingsolutions are given for problems of each stage you might encounter. In the datapreprocessing stage, complete the making up of missing values and the detection andtreatment of outlier. With stepwise regression analysis method to establish the initialmodel, and with analysis of variance to finish categorical factor analysis. After thecompletion of the modeling, proceed further testing and calibration for the model, toeliminate the multicollinearity and heteroscedasticity by ridge regression andweighted regression method between the impact factors, and improve the fit of themodel. Then, the introduction of the average error rates and expected rangeprobability is used to analyse the forecast performance of the cost regression model.Finally, the simulation is done for specific historical data sets. Build a software cost estimation model for the data set, specifically calculate the model performanceand predictive ability of all aspects during the regression modeling. The simulationresults show the feasibility of the model.